Quantification of continuous flood hazard using random forest classification and flood insurance claims at large spatial scales: a pilot study in southeast Texas

Abstract. Pre-disaster planning and mitigation necessitate detailed spatial information about flood hazards their associated risks. In the US, Federal Emergency Management Agency (FEMA) Special Flood Hazard Area (SFHA) provides important areas subject to flooding during 1 % riverine or coastal event. The binary nature of hazard maps obscures distribution property risk inside SFHA residual outside SFHA, which can undermine efforts. Machine learning techniques provide an alternative approach estimating across large scales at low computational expense. This study presents a pilot for Texas Gulf Coast region using random forest classification predict probability 30 523 km2 area. Using record National Insurance Program (NFIP) claims dating back 1976 high-resolution geospatial data, we generate continuous map 12 US Geological Survey (USGS) eight-digit hydrologic unit code (HUC) watersheds. Results indicate that model predicts with high sensitivity (area under curve, AUC: 0.895), especially compared existing FEMA regulatory floodplain. Our identifies 649 000 structures least annual chance flooding, roughly 3 times more than are currently identified by as flood-prone.